Machine learning for wearable IoT‐based applications: A survey

Abstract

This paper gives an overview about applying machine learning (ML) in wearable Wireless Body Area Network (WBAN). It highlights the main challenges and open issues for deploying ML models in such sensitive networks. The WBAN is an emerging technology in the last few years, which attracts lots of interest from the academic and industrial communities. It enables a wide range of IoT‐based applications in medical, lifestyle, sport, and entertainment. WBAN are constrained in many aspects such as those related to power resources, communication capabilities, and computation power. Moreover, these networks generate ample amount of sensory data. To overcome the limitations of these networks and make use of the big data generated, artificial intelligence is the best way to deal with gigantic data and help in automating several aspects of the network and its applications. This survey paper aims at reporting numerous ways ML is used to benefit these networks, the design factors that are considered when implementing the ML algorithms, and the communication technologies used in connecting wearable WBAN in the IoT era. The reported studies are based on real overviewed experiment and extensive simulation results.

Towards Energy Efficiency in the Internet of Wearable Things: A Systematic Review

“…In this regard, approximate computing or inexact computing have recently emerged as an effective technique where output accuracy is traded for computing time and energy by relying on nearly accurate results [211] (for a detailed discussion on approximate computing refer to Section IV-I.) Similarly, developing in-device signal processing and embedded ML techniques specifically designed for wearables has gained significant attention from the research community [212]. Most wearables are usually recording and communicating raw data to edge devices or remote cloud servers to be analyzed and processed for meaningful information extraction.…”

A novel holistic approach for performance evaluation in Internet of Things

“…The potential applications for IoT and LoRa/LoRaWAN are huge; 15 from e-health 16,17 and smart-grids 18,19 to intelligent transportation systems 20 and climate change. 21 For instance, low-cost, low-consumption nanosensors can be incorporated into clothes and accessories to monitor vital signs in patients; a mobile phone can act as a gateway to gather information and send it to a cloud system for further processing.…”

“…21 For instance, low-cost, low-consumption nanosensors can be incorporated into clothes and accessories to monitor vital signs in patients; a mobile phone can act as a gateway to gather information and send it to a cloud system for further processing. 16 Indeed, the performance of Wireless Body Area Networks (WBAN) could even be improved using LoRa in gateway-to-cloud communication, with the corresponding advantages. As another example, Al-Turjman and Abujubbeh 18 review the use of LoRa/LoRaWAN as an appropriate technology for neighborhood and wide area networks in smart grids, due to its IoT nature.…”

“…This propagation model uses several parameters to compute the FSPL, as shown in Table 6, where h b is the height of the LoRaGW antenna, h m is the height of the LoRa end node, f is the working frequency for this technology, and d m is the distance from a LoRa end node to the LoRaGW. The FSPL for three environments, namely, urban, suburban, and rural, can be computed with the Okumura-Hata model (15)(16)(17)(18). The model computes a correction factor a(h m ) that depends on LoRa end-node height and frequency and the chosen environment estimates L urban , L suburban or L rural .…”

Machine learning‐based edge‐computing on a multi‐level architecture of WSN and IoT for real‐time fall detection

“…For this reason, supervised learning is employed to distinguish various fall patterns from the ADL. The implementation of machine‐learning algorithms for online fall detection requires high performances in processing modules and power consumption [33], therefore, the wearable devices require to be lightweight as possible with minimised tightness, and a long battery lifetime [28]. To address these criteria, the learning of the classifier is done on a laptop using the Sisfall database [6], then, the actual test of obtained models is performed by embedding them in the Fog device for runtime experimentation.…”